mds: obviate MDLog::start_entry

The major problem here is that the MDLog::_start_entry method puts the
current event sequence number in the EMetaBlob of the event (if
present). Because of this, no other event can be submitted as this would
invalidate the event sequence. Instead, fixup the event sequence during
submission and simplify related logic that uses it during EMetaBlob
construction.

Secondarily, for the purposes of this commit series, _start_entry
introduced recursive locks when generating the ESubtreeMap within
MDLog::_segment_upkeep. So, this commit is a necessary cleanup.

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>

diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index e6cd03442bbd7c24d88b09d31ae1f2f726f7bf29..839d5fb65668554420b3cb5c0588f2a8e01de4d4 100644 (file)
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -2409,7 +2409,6 @@ void CInode::finish_scatter_update(ScatterLock *lock, CDir *dir,
       }
        
       EUpdate *le = new EUpdate(mdlog, ename);
-      mdlog->start_entry(le);
       le->metablob.add_dir_context(dir);
       le->metablob.add_dir(dir, true);
       

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 4e6930be47558b89613c22b01f92c6fdfa15cf04..557b8305941063fd93aab7299746f569666c71d2 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -2878,7 +2878,6 @@ bool Locker::check_inode_max_size(CInode *in, bool force_wrlock,
     metablob = &eu->metablob;
     le = eu;
   }
-  mds->mdlog->start_entry(le);
 
   mdcache->predirty_journal_parents(mut, metablob, in, 0, PREDIRTY_PRIMARY);
   // no cow, here!
@@ -2989,7 +2988,6 @@ void Locker::adjust_cap_wanted(Capability *cap, int wanted, int issue_seq)
     if (mdcache->open_file_table.should_log_open(cur)) {
       ceph_assert(cur->last == CEPH_NOSNAP);
       EOpen *le = new EOpen(mds->mdlog);
-      mds->mdlog->start_entry(le);
       le->add_clean_inode(cur);
       mds->mdlog->submit_entry(le);
     }
@@ -3603,7 +3601,6 @@ void Locker::_do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t foll
   }
 
   EUpdate *le = new EUpdate(mds->mdlog, "snap flush");
-  mds->mdlog->start_entry(le);
   MutationRef mut = new MutationImpl();
   mut->ls = mds->mdlog->get_current_segment();
 
@@ -3899,7 +3896,6 @@ bool Locker::_do_cap_update(CInode *in, Capability *cap,
 
   // do the update.
   EUpdate *le = new EUpdate(mds->mdlog, "cap update");
-  mds->mdlog->start_entry(le);
 
   bool xattr = (dirty & CEPH_CAP_XATTR_EXCL) &&
                m->xattrbl.length() &&
@@ -5005,7 +5001,6 @@ void Locker::scatter_writebehind(ScatterLock *lock)
   lock->start_flush();
 
   EUpdate *le = new EUpdate(mds->mdlog, "scatter_writebehind");
-  mds->mdlog->start_entry(le);
 
   mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mut.get(), &le->metablob, in);

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index d4ba1e53a474113bb940009cc7213514a1c157a5..f4ff185056a86d3d722a13ed5be0edae358293d4 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -532,7 +532,6 @@ void MDCache::_create_system_file(CDir *dir, std::string_view name, CInode *in,
 
   mut->ls = mds->mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mds->mdlog, "create system file");
-  mds->mdlog->start_entry(le);
 
   if (!in->is_mdsdir()) {
     predirty_journal_parents(mut, &le->metablob, in, dir, PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
@@ -2106,8 +2105,6 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
   bool do_parent_mtime = flags & PREDIRTY_DIR;
   bool shallow = flags & PREDIRTY_SHALLOW;
 
-  ceph_assert(mds->mdlog->entry_is_open());
-
   // make sure stamp is set
   if (mut->get_mds_stamp() == utime_t())
     mut->set_mds_stamp(ceph_clock_now());
@@ -2391,8 +2388,7 @@ void MDCache::log_leader_commit(metareqid_t reqid)
 {
   dout(10) << "log_leader_commit " << reqid << dendl;
   uncommitted_leaders[reqid].committing = true;
-  mds->mdlog->start_submit_entry(new ECommitted(reqid), 
-                                new C_MDC_CommittedLeader(this, reqid));
+  mds->mdlog->submit_entry(new ECommitted(reqid), new C_MDC_CommittedLeader(this, reqid));
 }
 
 void MDCache::_logged_leader_commit(metareqid_t reqid)
@@ -2509,7 +2505,6 @@ ESubtreeMap *MDCache::create_subtree_map()
   show_subtrees();
 
   ESubtreeMap *le = new ESubtreeMap();
-  mds->mdlog->_start_entry(le);
   
   map<dirfrag_t, CDir*> dirs_to_add;
 
@@ -3385,7 +3380,7 @@ void MDCache::handle_resolve_ack(const cref_t<MMDSResolveAck> &ack)
       ceph_assert(su);
 
       // log commit
-      mds->mdlog->start_submit_entry(new EPeerUpdate(mds->mdlog, "unknown", p.first, from,
+      mds->mdlog->submit_entry(new EPeerUpdate(mds->mdlog, "unknown", p.first, from,
                                                      EPeerUpdate::OP_COMMIT, su->origop),
                                     new C_MDC_PeerCommit(this, from, p.first));
       mds->mdlog->flush();
@@ -3598,7 +3593,7 @@ void MDCache::disambiguate_my_imports()
       dout(10) << "ambiguous import auth known, must not be me " << *dir << dendl;
       cancel_ambiguous_import(dir);
 
-      mds->mdlog->start_submit_entry(new EImportFinish(dir, false));
+      mds->mdlog->submit_entry(new EImportFinish(dir, false));
 
       // subtree may have been swallowed by another node claiming dir
       // as their own.
@@ -3610,7 +3605,7 @@ void MDCache::disambiguate_my_imports()
     } else {
       dout(10) << "ambiguous import auth unclaimed, must be me " << *dir << dendl;
       finish_ambiguous_import(q->first);
-      mds->mdlog->start_submit_entry(new EImportFinish(dir, true));
+      mds->mdlog->submit_entry(new EImportFinish(dir, true));
     }
   }
   ceph_assert(my_ambiguous_imports.empty());
@@ -5438,7 +5433,7 @@ bool MDCache::process_imported_caps()
                                                              finish->session_map);
       ESessions *le = new ESessions(pv, std::move(rejoin_client_map),
                                    std::move(rejoin_client_metadata_map));
-      mds->mdlog->start_submit_entry(le, finish);
+      mds->mdlog->submit_entry(le, finish);
       mds->mdlog->flush();
       rejoin_client_map.clear();
       rejoin_client_metadata_map.clear();
@@ -6288,7 +6283,6 @@ void MDCache::queue_file_recover(CInode *in)
     auto mut(std::make_shared<MutationImpl>());
     mut->ls = mds->mdlog->get_current_segment();
     EUpdate *le = new EUpdate(mds->mdlog, "queue_file_recover cow");
-    mds->mdlog->start_entry(le);
     predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY);
 
     s.erase(*s.begin());
@@ -6634,7 +6628,6 @@ void MDCache::truncate_inode_finish(CInode *in, LogSegment *ls)
   pi.inode->fscrypt_last_block = bufferlist();
 
   EUpdate *le = new EUpdate(mds->mdlog, "truncate finish");
-  mds->mdlog->start_entry(le);
 
   predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY);
   journal_dirty_inode(mut.get(), &le->metablob, in);
@@ -6748,7 +6741,7 @@ void MDCache::purge_inodes(const interval_set<inodeno_t>& inos, LogSegment *ls)
       mds->inotable->project_release_ids(inos);
       version_t piv = mds->inotable->get_projected_version();
       ceph_assert(piv != 0);
-      mds->mdlog->start_submit_entry(new EPurged(inos, ls->seq, piv),
+      mds->mdlog->submit_entry(new EPurged(inos, ls->seq, piv),
                                     new C_MDS_purge_completed_finish(this, inos, ls, piv));
       mds->mdlog->flush();
     });
@@ -7959,7 +7952,8 @@ bool MDCache::shutdown_pass()
     if (ls->num_events > 1 || !ls->dirty_dirfrags.empty()) {
       // Current segment contains events other than subtreemap or
       // there are dirty dirfrags (see CDir::log_mark_dirty())
-      mds->mdlog->start_new_segment();
+      auto sle = create_subtree_map();
+      mds->mdlog->submit_entry(sle);
       mds->mdlog->flush();
     }
   }
@@ -12066,7 +12060,6 @@ void MDCache::dispatch_fragment_dir(MDRequestRef& mdr)
 
   mdr->ls = mds->mdlog->get_current_segment();
   EFragment *le = new EFragment(mds->mdlog, EFragment::OP_PREPARE, basedirfrag, info.bits);
-  mds->mdlog->start_entry(le);
 
   for (const auto& dir : info.dirs) {
     dirfrag_rollback rollback;
@@ -12210,7 +12203,7 @@ void MDCache::_fragment_stored(MDRequestRef& mdr)
 
   // journal commit
   EFragment *le = new EFragment(mds->mdlog, EFragment::OP_COMMIT, basedirfrag, info.bits);
-  mds->mdlog->start_submit_entry(le, new C_MDC_FragmentCommit(this, basedirfrag, mdr));
+  mds->mdlog->submit_entry(le, new C_MDC_FragmentCommit(this, basedirfrag, mdr));
 
 
   // unfreeze resulting frags
@@ -12280,7 +12273,7 @@ void MDCache::_fragment_old_purged(dirfrag_t basedirfrag, int bits, const MDRequ
     mdr->mark_event("old frags purged");
 
   EFragment *le = new EFragment(mds->mdlog, EFragment::OP_FINISH, basedirfrag, bits);
-  mds->mdlog->start_submit_entry(le);
+  mds->mdlog->submit_entry(le);
 
   finish_uncommitted_fragment(basedirfrag, EFragment::OP_FINISH);
 
@@ -12495,7 +12488,6 @@ void MDCache::rollback_uncommitted_fragments()
     MutationRef mut(new MutationImpl());
     mut->ls = mds->mdlog->get_current_segment();
     EFragment *le = new EFragment(mds->mdlog, EFragment::OP_ROLLBACK, p->first, uf.bits);
-    mds->mdlog->start_entry(le);
     bool diri_auth = (diri->authority() != CDIR_AUTH_UNDEF);
 
     frag_vec_t old_frags;
@@ -13181,7 +13173,6 @@ void MDCache::repair_dirfrag_stats_work(MDRequestRef& mdr)
 
   mdr->ls = mds->mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mds->mdlog, "repair_dirfrag");
-  mds->mdlog->start_entry(le);
 
   if (!good_fragstat) {
     if (pf->fragstat.mtime > frag_info.mtime)

diff --git a/src/mds/MDLog.cc b/src/mds/MDLog.cc
index 5e1540862d3023e52bd9ef7c8ef615934db86546..c131da5e4f0da3b9abd68f854fdb162413f51417 100644 (file)
--- a/src/mds/MDLog.cc
+++ b/src/mds/MDLog.cc
@@ -267,37 +267,22 @@ void MDLog::append()
 
 // -------------------------------------------------
 
-void MDLog::_start_entry(LogEvent *e)
+void MDLog::_submit_entry(LogEvent *le, MDSLogContextBase* c)
 {
+  dout(20) << __func__ << " " << *le << dendl;
+  ceph_assert(ceph_mutex_is_locked_by_me(mds->mds_lock));
   ceph_assert(ceph_mutex_is_locked_by_me(submit_mutex));
-
-  ceph_assert(cur_event == NULL);
-  cur_event = e;
+  ceph_assert(!mds->is_any_replay());
+  ceph_assert(!mds_is_shutting_down);
 
   event_seq++;
 
-  EMetaBlob *metablob = e->get_metablob();
+  EMetaBlob *metablob = le->get_metablob();
   if (metablob) {
-    metablob->event_seq = event_seq;
-    metablob->last_subtree_map = get_last_segment_seq();
+    for (auto& in : metablob->get_touched_inodes()) {
+      in->last_journaled = event_seq;
+    }
   }
-}
-
-void MDLog::cancel_entry(LogEvent *le)
-{
-  ceph_assert(le == cur_event);
-  cur_event = NULL;
-  delete le;
-}
-
-void MDLog::_submit_entry(LogEvent *le, MDSLogContextBase *c)
-{
-  ceph_assert(ceph_mutex_is_locked_by_me(submit_mutex));
-  ceph_assert(!mds->is_any_replay());
-  ceph_assert(!mds_is_shutting_down);
-
-  ceph_assert(le == cur_event);
-  cur_event = NULL;
 
   // let the event register itself in the segment
   ceph_assert(!segments.empty());
@@ -318,8 +303,14 @@ void MDLog::_submit_entry(LogEvent *le, MDSLogContextBase *c)
   }
 
   unflushed++;
-  
+}
+
+void MDLog::_segment_upkeep(LogEvent* le)
+{
+  ceph_assert(ceph_mutex_is_locked_by_me(mds->mds_lock));
+  ceph_assert(ceph_mutex_is_locked_by_me(submit_mutex));
   uint64_t period = journaler->get_layout_period();
+  auto ls = get_current_segment();
   // start a new segment?
   if (le->get_type() == EVENT_SUBTREEMAP ||
       (le->get_type() == EVENT_IMPORTFINISH && mds->is_resolve())) {
@@ -327,16 +318,15 @@ void MDLog::_submit_entry(LogEvent *le, MDSLogContextBase *c)
     // do not insert ESubtreeMap among EImportFinish events that finish
     // disambiguate imports. Because the ESubtreeMap reflects the subtree
     // state when all EImportFinish events are replayed.
-  } else if (ls->end/period != ls->offset/period ||
-            ls->num_events >= events_per_segment) {
+  } else if (ls->end/period != ls->offset/period || ls->num_events >= events_per_segment) {
     dout(10) << "submit_entry also starting new segment: last = "
-            << ls->seq  << "/" << ls->offset << ", event seq = " << event_seq << dendl;
+            << ls->seq  << "/" << ls->offset << ", event seq = " << event_seq << dendl;
     _start_new_segment();
-  } else if (debug_subtrees &&
-            le->get_type() != EVENT_SUBTREEMAP_TEST) {
+  } else if (debug_subtrees && le->get_type() != EVENT_SUBTREEMAP_TEST) {
     // debug: journal this every time to catch subtree replay bugs.
     // use a different event id so it doesn't get interpreted as a
     // LogSegment boundary on replay.
+    dout(10) << __func__ << ": creating test subtree map" << dendl;
     LogEvent *sle = mds->mdcache->create_subtree_map();
     sle->set_type(EVENT_SUBTREEMAP_TEST);
     _submit_entry(sle, NULL);

diff --git a/src/mds/MDLog.h b/src/mds/MDLog.h
index d5e7c5a8a7df2496eba266172f60c94df6c1e4a5..0e3d26687a2acc5bbc7764b7ad3f7c54d65f2643 100644 (file)
--- a/src/mds/MDLog.h
+++ b/src/mds/MDLog.h
@@ -130,31 +130,23 @@ public:
   Journaler *get_journaler() { return journaler; }
   bool empty() const { return segments.empty(); }
 
+  uint64_t get_last_segment_seq() const {
+    ceph_assert(!segments.empty());
+    return segments.rbegin()->first;
+  }
+
   bool is_capped() const { return mds_is_shutting_down; }
   void cap();
 
   void kick_submitter();
   void shutdown();
 
-  void _start_entry(LogEvent *e);
-  void start_entry(LogEvent *e) {
-    std::lock_guard l(submit_mutex);
-    _start_entry(e);
-  }
-  void cancel_entry(LogEvent *e);
-  void _submit_entry(LogEvent *e, MDSLogContextBase *c);
-  void submit_entry(LogEvent *e, MDSLogContextBase *c = 0) {
-    std::lock_guard l(submit_mutex);
-    _submit_entry(e, c);
-    submit_cond.notify_all();
-  }
-  void start_submit_entry(LogEvent *e, MDSLogContextBase *c = 0) {
+  void submit_entry(LogEvent *e, MDSLogContextBase* c = 0) {
     std::lock_guard l(submit_mutex);
-    _start_entry(e);
     _submit_entry(e, c);
+    _segment_upkeep(e);
     submit_cond.notify_all();
   }
-  bool entry_is_open() const { return cur_event != NULL; }
 
   void wait_for_safe(Context* c);
   void flush();
@@ -250,10 +242,6 @@ protected:
 
   void _submit_thread();
 
-  uint64_t get_last_segment_seq() const {
-    ceph_assert(!segments.empty());
-    return segments.rbegin()->first;
-  }
   LogSegment *get_oldest_segment() {
     return segments.begin()->second;
   }
@@ -303,7 +291,9 @@ private:
   // -- segments --
   void _start_new_segment();
   void _prepare_new_segment();
+  void _segment_upkeep(LogEvent* le);
   void _journal_segment_subtree_map(MDSContext *onsync);
+  void _submit_entry(LogEvent *e, MDSLogContextBase *c);
 
   void try_to_commit_open_file_table(uint64_t last_seq);
 
@@ -313,9 +303,6 @@ private:
   void _trim_expired_segments();
   void write_head(MDSContext *onfinish);
 
-  // -- events --
-  LogEvent *cur_event = nullptr;
-
   bool debug_subtrees;
   uint64_t events_per_segment;
   int64_t max_events;

diff --git a/src/mds/MDSTableClient.cc b/src/mds/MDSTableClient.cc
index 9ded20bb8a86a017dd9d741aeb96437e0137d837..fc8103ee992899a74864799b0fed03137ce3163b 100644 (file)
--- a/src/mds/MDSTableClient.cc
+++ b/src/mds/MDSTableClient.cc
@@ -117,7 +117,7 @@ void MDSTableClient::handle_request(const cref_t<MMDSTableRequest> &m)
       pending_commit.erase(tid);
 
       // log ACK.
-      mds->mdlog->start_submit_entry(new ETableClient(table, TABLESERVER_OP_ACK, tid),
+      mds->mdlog->submit_entry(new ETableClient(table, TABLESERVER_OP_ACK, tid),
                                     new C_LoggedAck(this, tid));
     } else {
       dout(10) << "got stray ack on tid " << tid << ", ignoring" << dendl;

diff --git a/src/mds/MDSTableServer.cc b/src/mds/MDSTableServer.cc
index 6e542b4e396560091342fc64cb48689ab55636b8..5514f1e784784909e5e7fa0dbbb64cdf2dbe00b7 100644 (file)
--- a/src/mds/MDSTableServer.cc
+++ b/src/mds/MDSTableServer.cc
@@ -64,7 +64,6 @@ void MDSTableServer::handle_prepare(const cref_t<MMDSTableRequest> &req)
 
   ETableServer *le = new ETableServer(table, TABLESERVER_OP_PREPARE, req->reqid, from,
                                      projected_version, projected_version);
-  mds->mdlog->start_entry(le);
   le->mutation = req->bl;
   mds->mdlog->submit_entry(le, new C_Prepare(this, req, projected_version));
   mds->mdlog->flush();
@@ -148,7 +147,7 @@ void MDSTableServer::handle_commit(const cref_t<MMDSTableRequest> &req)
     projected_version++;
     committing_tids.insert(tid);
 
-    mds->mdlog->start_submit_entry(new ETableServer(table, TABLESERVER_OP_COMMIT, 0, MDS_RANK_NONE, 
+    mds->mdlog->submit_entry(new ETableServer(table, TABLESERVER_OP_COMMIT, 0, MDS_RANK_NONE, 
                                                    tid, projected_version),
                                   new C_Commit(this, req));
   }
@@ -206,7 +205,7 @@ void MDSTableServer::handle_rollback(const cref_t<MMDSTableRequest> &req)
   projected_version++;
   committing_tids.insert(tid);
 
-  mds->mdlog->start_submit_entry(new ETableServer(table, TABLESERVER_OP_ROLLBACK, 0, MDS_RANK_NONE,
+  mds->mdlog->submit_entry(new ETableServer(table, TABLESERVER_OP_ROLLBACK, 0, MDS_RANK_NONE,
                                                  tid, projected_version),
                                 new C_Rollback(this, req));
 }
@@ -245,7 +244,6 @@ void MDSTableServer::do_server_update(bufferlist& bl)
   projected_version++;
 
   ETableServer *le = new ETableServer(table, TABLESERVER_OP_SERVER_UPDATE, 0, MDS_RANK_NONE, 0, projected_version);
-  mds->mdlog->start_entry(le);
   le->mutation = bl;
   mds->mdlog->submit_entry(le, new C_ServerUpdate(this, bl));
 }

diff --git a/src/mds/Migrator.cc b/src/mds/Migrator.cc
index 8bd875c3450ef89c4ad7d948bf3c27a2efc570fc..1dd4cb17737ef13fa1c9b459ae1c449a43dcd892 100644 (file)
--- a/src/mds/Migrator.cc
+++ b/src/mds/Migrator.cc
@@ -1905,7 +1905,6 @@ void Migrator::handle_export_ack(const cref_t<MExportDirAck> &m)
   // log completion. 
   //  include export bounds, to ensure they're in the journal.
   EExport *le = new EExport(mds->mdlog, dir, it->second.peer);;
-  mds->mdlog->start_entry(le);
 
   le->metablob.add_dir_context(dir, EMetaBlob::TO_ROOT);
   le->metablob.add_dir(dir, false);
@@ -2670,7 +2669,6 @@ void Migrator::handle_export_dir(const cref_t<MExportDir> &m)
 
   // start the journal entry
   EImportStart *le = new EImportStart(mds->mdlog, dir->dirfrag(), m->bounds, oldauth);
-  mds->mdlog->start_entry(le);
 
   le->metablob.add_dir_context(dir);
   
@@ -2910,7 +2908,7 @@ void Migrator::import_reverse(CDir *dir)
   }
         
   // log our failure
-  mds->mdlog->start_submit_entry(new EImportFinish(dir, false));       // log failure
+  mds->mdlog->submit_entry(new EImportFinish(dir, false));     // log failure
 
   mdcache->trim(num_dentries); // try trimming dentries
 
@@ -3129,7 +3127,7 @@ void Migrator::import_finish(CDir *dir, bool notify, bool last)
   MutationRef mut = it->second.mut;
   import_state.erase(it);
 
-  mds->mdlog->start_submit_entry(new EImportFinish(dir, true));
+  mds->mdlog->submit_entry(new EImportFinish(dir, true));
 
   // process delayed expires
   mdcache->process_delayed_expire(dir);
@@ -3638,7 +3636,7 @@ void Migrator::handle_export_caps(const cref_t<MExportCaps> &ex)
   // journal open client sessions
   ESessions *le = new ESessions(pv, std::move(client_map),
                                std::move(client_metadata_map));
-  mds->mdlog->start_submit_entry(le, finish);
+  mds->mdlog->submit_entry(le, finish);
   mds->mdlog->flush();
 }
 

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 6d91d5fcda227c6b151b0fd908df8c3a60a692e7..91dd0a3bd7d623755b1f479a4ed1d1009b3fc4df 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -759,7 +759,7 @@ void Server::handle_client_session(const cref_t<MClientSession> &m)
         ceph_assert(r == 0);
         log_session_status("ACCEPTED", "");
       });
-      mdlog->start_submit_entry(new ESession(m->get_source_inst(), true, pv, client_metadata),
+      mdlog->submit_entry(new ESession(m->get_source_inst(), true, pv, client_metadata),
                                new C_MDS_session_finish(this, session, sseq, true, pv, fin));
       mdlog->flush();
     }
@@ -1438,7 +1438,7 @@ void Server::journal_close_session(Session *session, int state, Context *on_safe
   auto le = new ESession(session->info.inst, false, pv, inos_to_free, piv, session->delegated_inos);
   auto fin = new C_MDS_session_finish(this, session, sseq, false, pv, inos_to_free, piv,
                                      session->delegated_inos, mdlog->get_current_segment(), on_safe);
-  mdlog->start_submit_entry(le, fin);
+  mdlog->submit_entry(le, fin);
   mdlog->flush();
 
   // clean up requests, too
@@ -4525,7 +4525,6 @@ void Server::handle_client_open(MDRequestRef& mdr)
   if (cur->is_auth() && cur->last == CEPH_NOSNAP &&
       mdcache->open_file_table.should_log_open(cur)) {
     EOpen *le = new EOpen(mds->mdlog);
-    mdlog->start_entry(le);
     le->add_clean_inode(cur);
     mdlog->submit_entry(le);
   }
@@ -4727,7 +4726,6 @@ void Server::handle_client_openc(MDRequestRef& mdr)
   // prepare finisher
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "openc");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
@@ -5366,7 +5364,6 @@ void Server::handle_client_setattr(MDRequestRef& mdr)
   // project update
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setattr");
-  mdlog->start_entry(le);
 
   auto pi = cur->project_inode(mdr);
 
@@ -5458,7 +5455,6 @@ void Server::do_open_truncate(MDRequestRef& mdr, int cmode)
 
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "open_truncate");
-  mdlog->start_entry(le);
 
   // prepare
   auto pi = in->project_inode(mdr);
@@ -5591,7 +5587,6 @@ void Server::handle_client_setlayout(MDRequestRef& mdr)
   // log + wait
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setlayout");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -5713,7 +5708,6 @@ void Server::handle_client_setdirlayout(MDRequestRef& mdr)
   // log + wait
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setlayout");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -6307,7 +6301,6 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur)
   // log + wait
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "set vxattr layout");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -6352,7 +6345,6 @@ void Server::handle_remove_vxattr(MDRequestRef& mdr, CInode *cur)
     // log + wait
     mdr->ls = mdlog->get_current_segment();
     EUpdate *le = new EUpdate(mdlog, "remove dir layout vxattr");
-    mdlog->start_entry(le);
     le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
     mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
     mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -6641,7 +6633,6 @@ void Server::handle_client_setxattr(MDRequestRef& mdr)
   // log + wait
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "setxattr");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -6711,7 +6702,6 @@ void Server::handle_client_removexattr(MDRequestRef& mdr)
   // log + wait
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "removexattr");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
@@ -7101,7 +7091,6 @@ void Server::handle_client_mknod(MDRequestRef& mdr)
   // prepare finisher
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "mknod");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   
@@ -7184,7 +7173,6 @@ void Server::handle_client_mkdir(MDRequestRef& mdr)
   // prepare finisher
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "mkdir");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
@@ -7269,7 +7257,6 @@ void Server::handle_client_symlink(MDRequestRef& mdr)
   // prepare finisher
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "symlink");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   journal_allocated_inos(mdr, &le->metablob);
   mdcache->predirty_journal_parents(mdr, &le->metablob, newi, dn->get_dir(), PREDIRTY_PRIMARY|PREDIRTY_DIR, 1);
@@ -7472,7 +7459,6 @@ void Server::_link_local(MDRequestRef& mdr, CDentry *dn, CInode *targeti, SnapRe
 
   // log + wait
   EUpdate *le = new EUpdate(mdlog, "link_local");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, dn->get_dir(), PREDIRTY_DIR, 1);      // new dn
   mdcache->predirty_journal_parents(mdr, &le->metablob, targeti, 0, PREDIRTY_PRIMARY);           // targeti
@@ -7583,7 +7569,6 @@ void Server::_link_remote(MDRequestRef& mdr, bool inc, CDentry *dn, CInode *targ
   // add to event
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, inc ? "link_remote":"unlink_remote");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   if (!mdr->more()->witnessed.empty()) {
     dout(20) << " noting uncommitted_peers " << mdr->more()->witnessed << dendl;
@@ -7713,8 +7698,6 @@ void Server::handle_peer_link_prep(MDRequestRef& mdr)
   mdr->ls = mdlog->get_current_segment();
   EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_link_prep", mdr->reqid, mdr->peer_to_mds,
                                      EPeerUpdate::OP_PREPARE, EPeerUpdate::LINK);
-  mdlog->start_entry(le);
-
   auto pi = dnl->get_inode()->project_inode(mdr);
 
   // update journaled target inode
@@ -7846,7 +7829,6 @@ void Server::_commit_peer_link(MDRequestRef& mdr, int r, CInode *targeti)
     // write a commit to the journal
     EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_link_commit", mdr->reqid, mdr->peer_to_mds,
                                        EPeerUpdate::OP_COMMIT, EPeerUpdate::LINK);
-    mdlog->start_entry(le);
     submit_mdlog_entry(le, new C_MDS_CommittedPeer(this, mdr), mdr, __func__);
     mdlog->flush();
   } else {
@@ -7949,7 +7931,6 @@ void Server::do_link_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef&
   // journal it
   EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_link_rollback", rollback.reqid, leader,
                                      EPeerUpdate::OP_ROLLBACK, EPeerUpdate::LINK);
-  mdlog->start_entry(le);
   le->commit.add_dir_context(parent);
   le->commit.add_dir(parent, true);
   le->commit.add_primary_dentry(in->get_projected_parent_dn(), 0, true);
@@ -8212,7 +8193,6 @@ void Server::_unlink_local(MDRequestRef& mdr, CDentry *dn, CDentry *straydn)
 
   // prepare log entry
   EUpdate *le = new EUpdate(mdlog, "unlink_local");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(mdr->reqid, mdr->client_request->get_oldest_client_tid());
   if (!mdr->more()->witnessed.empty()) {
     dout(20) << " noting uncommitted_peers " << mdr->more()->witnessed << dendl;
@@ -8470,7 +8450,6 @@ void Server::handle_peer_rmdir_prep(MDRequestRef& mdr)
   mdr->ls = mdlog->get_current_segment();
   EPeerUpdate *le =  new EPeerUpdate(mdlog, "peer_rmdir", mdr->reqid, mdr->peer_to_mds,
                                       EPeerUpdate::OP_PREPARE, EPeerUpdate::RMDIR);
-  mdlog->start_entry(le);
   le->rollback = mdr->more()->rollback_bl;
 
   le->commit.add_dir_context(straydn->get_dir());
@@ -8569,7 +8548,6 @@ void Server::_commit_peer_rmdir(MDRequestRef& mdr, int r, CDentry *straydn)
       EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_rmdir_commit", mdr->reqid,
                                          mdr->peer_to_mds, EPeerUpdate::OP_COMMIT,
                                          EPeerUpdate::RMDIR);
-      mdlog->start_entry(le);
       submit_mdlog_entry(le, new C_MDS_CommittedPeer(this, mdr), mdr, __func__);
       mdlog->flush();
     } else {
@@ -8645,7 +8623,6 @@ void Server::do_rmdir_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef&
 
   EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_rmdir_rollback", rollback.reqid, leader,
                                      EPeerUpdate::OP_ROLLBACK, EPeerUpdate::RMDIR);
-  mdlog->start_entry(le);
   
   le->commit.add_dir_context(dn->get_dir());
   le->commit.add_primary_dentry(dn, in, true);
@@ -9207,7 +9184,6 @@ void Server::handle_client_rename(MDRequestRef& mdr)
   // -- prepare journal entry --
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "rename");
-  mdlog->start_entry(le);
   le->metablob.add_client_req(mdr->reqid, req->get_oldest_client_tid());
   if (!mdr->more()->witnessed.empty()) {
     dout(20) << " noting uncommitted_peers " << mdr->more()->witnessed << dendl;
@@ -10213,7 +10189,6 @@ void Server::handle_peer_rename_prep(MDRequestRef& mdr)
   mdr->ls = mdlog->get_current_segment();
   EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_rename_prep", mdr->reqid, mdr->peer_to_mds,
                                      EPeerUpdate::OP_PREPARE, EPeerUpdate::RENAME);
-  mdlog->start_entry(le);
   le->rollback = mdr->more()->rollback_bl;
   
   bufferlist blah;  // inode import data... obviously not used if we're the peer
@@ -10221,7 +10196,7 @@ void Server::handle_peer_rename_prep(MDRequestRef& mdr)
 
   if (le->commit.empty()) {
     dout(10) << " empty metablob, skipping journal" << dendl;
-    mdlog->cancel_entry(le);
+    delete le;
     mdr->ls = NULL;
     _logged_peer_rename(mdr, srcdn, destdn, straydn);
   } else {
@@ -10375,7 +10350,6 @@ void Server::_commit_peer_rename(MDRequestRef& mdr, int r,
       EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_rename_commit", mdr->reqid,
                                          mdr->peer_to_mds, EPeerUpdate::OP_COMMIT,
                                          EPeerUpdate::RENAME);
-      mdlog->start_entry(le);
       submit_mdlog_entry(le, new C_MDS_CommittedPeer(this, mdr), mdr, __func__);
       mdlog->flush();
     } else {
@@ -10719,7 +10693,6 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef
   // journal it
   EPeerUpdate *le = new EPeerUpdate(mdlog, "peer_rename_rollback", rollback.reqid, leader,
                                      EPeerUpdate::OP_ROLLBACK, EPeerUpdate::RENAME);
-  mdlog->start_entry(le);
 
   if (srcdn && (srcdn->authority().first == whoami || force_journal_src)) {
     le->commit.add_dir_context(srcdir);
@@ -10777,7 +10750,7 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t leader, MDRequestRef
 
   if (mdr && !mdr->more()->peer_update_journaled) {
     ceph_assert(le->commit.empty());
-    mdlog->cancel_entry(le);
+    delete le;
     mut->ls = NULL;
     _rename_rollback_finish(mut, mdr, srcdn, srcdnpv, destdn, straydn, splits, finish_mdr);
   } else {
@@ -11207,7 +11180,6 @@ void Server::handle_client_mksnap(MDRequestRef& mdr)
   // journal the inode changes
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "mksnap");
-  mdlog->start_entry(le);
 
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   le->metablob.add_table_transaction(TABLE_SNAP, stid);
@@ -11333,7 +11305,6 @@ void Server::handle_client_rmsnap(MDRequestRef& mdr)
   
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "rmsnap");
-  mdlog->start_entry(le);
   
   // project the snaprealm
   auto &newnode = *pi.snapnode;
@@ -11485,7 +11456,6 @@ void Server::handle_client_renamesnap(MDRequestRef& mdr)
   // journal the inode changes
   mdr->ls = mdlog->get_current_segment();
   EUpdate *le = new EUpdate(mdlog, "renamesnap");
-  mdlog->start_entry(le);
 
   le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
   le->metablob.add_table_transaction(TABLE_SNAP, stid);

diff --git a/src/mds/StrayManager.cc b/src/mds/StrayManager.cc
index d5827d48002c55b179818ab67002ec3340e76774..ec63b1d48b810b20a72018e96ad43a688e7f4195 100644 (file)
--- a/src/mds/StrayManager.cc
+++ b/src/mds/StrayManager.cc
@@ -202,7 +202,6 @@ void StrayManager::_purge_stray_purged(
     pf->version = dir->pre_dirty();
 
     EUpdate *le = new EUpdate(mds->mdlog, "purge_stray truncate");
-    mds->mdlog->start_entry(le);
 
     le->metablob.add_dir_context(dir);
     auto& dl = le->metablob.add_dir(dn->dir, true);
@@ -230,7 +229,6 @@ void StrayManager::_purge_stray_purged(
     dn->push_projected_linkage(); // NULL
 
     EUpdate *le = new EUpdate(mds->mdlog, "purge_stray");
-    mds->mdlog->start_entry(le);
 
     // update dirfrag fragstat, rstat
     CDir *dir = dn->get_dir();

diff --git a/src/mds/events/EMetaBlob.h b/src/mds/events/EMetaBlob.h
index 736a509ea211dbf0bc474b313394b9e32e896efe..6c995dddeb029ef81dbad0d551eb80590320e9b2 100644 (file)
--- a/src/mds/events/EMetaBlob.h
+++ b/src/mds/events/EMetaBlob.h
@@ -340,26 +340,27 @@ private:
   std::vector<std::pair<metareqid_t,uint64_t> > client_reqs;
   std::vector<std::pair<metareqid_t,uint64_t> > client_flushes;
 
+  std::set<CInode*> touched;
+
  public:
   void encode(bufferlist& bl, uint64_t features) const;
   void decode(bufferlist::const_iterator& bl);
   void get_inodes(std::set<inodeno_t> &inodes) const;
+  const auto& get_touched_inodes(void) const {
+    return touched;
+  }
   void get_paths(std::vector<std::string> &paths) const;
   void get_dentries(std::map<dirfrag_t, std::set<std::string> > &dentries) const;
   entity_name_t get_client_name() const {return client_name;}
 
   void dump(Formatter *f) const;
   static void generate_test_instances(std::list<EMetaBlob*>& ls);
-  // soft stateadd
-  uint64_t last_subtree_map;
-  uint64_t event_seq;
 
   // for replay, in certain cases
   //LogSegment *_segment;
 
   EMetaBlob() : opened_ino(0), renamed_dirino(0),
-                inotablev(0), sessionmapv(0), allocated_ino(0),
-                last_subtree_map(0), event_seq(0)
+                inotablev(0), sessionmapv(0), allocated_ino(0)
                 {}
   EMetaBlob(const EMetaBlob&) = delete;
   ~EMetaBlob() { }
@@ -480,7 +481,8 @@ private:
                    state, in->get_old_inodes());
 
     // make note of where this inode was last journaled
-    in->last_journaled = event_seq;
+
+    touched.insert(in);
     //cout << "journaling " << in->inode.ino << " at " << my_offset << std::endl;
   }
 
@@ -512,7 +514,7 @@ private:
   }
 
   void add_root(bool dirty, CInode *in) {
-    in->last_journaled = event_seq;
+    touched.insert(in);
     //cout << "journaling " << in->inode.ino << " at " << my_offset << std::endl;
 
     const auto& pi = in->get_projected_inode();

diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index 96e5295ef35da7471bbc26240387bfbb6cb0e468..f4fccad2e5c0b0c5c10e04c33abcd7d54c5985ec 100644 (file)
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -218,7 +218,6 @@ void LogSegment::try_to_expire(MDSRank *mds, MDSGatherBuilder &gather_bld, int o
        dout(20) << "try_to_expire requeueing snap needflush inode " << *in << dendl;
        if (!le) {
          le = new EOpen(mds->mdlog);
-         mds->mdlog->start_entry(le);
        }
        le->add_clean_inode(in);
        ls->open_files.push_back(&in->item_open_file);
@@ -391,15 +390,16 @@ void EMetaBlob::add_dir_context(CDir *dir, int mode)
       }
 
       // was the inode journaled in this blob?
-      if (event_seq && diri->last_journaled == event_seq) {
+      if (touched.contains(diri)) {
        dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri this blob " << *diri << dendl;
        break;
       }
 
       // have we journaled this inode since the last subtree map?
-      if (!maybenot && last_subtree_map && diri->last_journaled >= last_subtree_map) {
+      auto last_segment_seq = mds->mdlog->get_last_segment_seq();
+      if (!maybenot && diri->last_journaled >= last_segment_seq) {
        dout(20) << "EMetaBlob::add_dir_context(" << dir << ") already have diri in this segment (" 
-                << diri->last_journaled << " >= " << last_subtree_map << "), setting maybenot flag "
+                << diri->last_journaled << " >= " << last_segment_seq << "), setting maybenot flag "
                 << *diri << dendl;
        maybenot = true;
       }